Summary of Confidence-aware Contrastive Learning For Selective Classification, by Yu-chang Wu et al.
Confidence-aware Contrastive Learning for Selective Classification
by Yu-Chang Wu, Shen-Huan Lyu, Haopu Shang, Xiangyu Wang, Chao Qian
First submitted to arxiv on: 7 Jun 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Computer Vision and Pattern Recognition (cs.CV)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary Selective classification enables models to make predictions only when they are sufficiently confident, which is crucial in high-stakes scenarios. Previous methods focus on modifying deep neural networks’ architecture to estimate prediction confidence. This work provides a generalization bound for selective classification, revealing that optimizing feature layers improves performance. Inspired by this theory, we propose Confidence-aware Contrastive Learning method (CCL-SC) for Selective Classification, which similarizes homogeneous instance features and differentiates heterogeneous instance features based on the model’s confidence strength. Experimental results on CIFAR-10, CIFAR-100, CelebA, and ImageNet datasets show that CCL-SC achieves lower selective risk than state-of-the-art methods across most coverage degrees. Moreover, it can be combined with existing methods for further improvement. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary Imagine a smart model that only makes predictions when it’s really sure about what it’s saying. This is important in situations where you need to get things right the first time. Previous attempts at making these models work better have focused on changing how they think, rather than how they look. Now, we’re taking a different approach by improving the way our model processes information (its “features”). We’ve come up with a new method called Confidence-aware Contrastive Learning for Selective Classification (CCL-SC). It helps our model tell apart similar and very different things, based on how sure it is about its predictions. We tested this method on some big datasets and found that it performs better than other methods in many cases. It’s also flexible enough to be used with existing methods to get even better results. |
Keywords
* Artificial intelligence * Classification * Generalization
